Lecture17

advertisement
Engineering Analysis ENG 3420 Fall 2009
Dan C. Marinescu
Office: HEC 439 B
Office hours: Tu-Th 11:00-12:00
Lecture 17


Reading assignment Chapters 10 and 11, Linear Algebra ClassNotes
Last time:



Today:








Symmetric matrices; Hermitian matrices.
Matrix multiplication
Linear algebra functions in Matlab
The inverse of a matrix
Vector products
Tensor algebra
Characteristic equation, eigenvectors, eigenvalues
Norm
Matrix condition number
Next Time

More on


LU Factorization
Cholesky decomposition
Lecture 17
2
Matrix analysis in MATLAB
Norm
normest
rank
det
trace
null
orth
rref
subspace
 Matrix or vector norm
 Estimate the matrix 2-norm
 Matrix rank
 Determinant
 Sum of diagonal elements
 Null space
 Orthogonalization
 Reduced row echelon form
Angle between two subspaces
Eigenvalues and singular values
eig
svd
eigs
svds
poly
polyeig
condeig
hess
qz
schur
 Eigenvalues and eigenvectors
 Singular value decomposition
 A few eigenvalues
 A few singular values
 Characteristic polynomial
 Polynomial eigenvalue problem
 Condition number for eigenvalues
 Hessenberg form
 QZ factorization
 Schur decomposition
Matrix functions
Expm
Logm
Sqrtm
Funm
 Matrix exponential
 Matrix logarithm
 Matrix square root
 Evaluate general matrix function
Linear systems of equations
\ and /
 Linear equation solution
inv
 Matrix inverse
cond
 Condition number for inversion
condest  1-norm condition number estimate
chol
 Cholesky factorization
cholinc
 Incomplete Cholesky factorization
linsolve  Solve a system of linear equations
lu
 LU factorization
ilu
 Incomplete LU factorization
luinc
 Incomplete LU factorization
qr
 Orthogonal-triangular decomposition
lsqnonneg  Nonnegative least-squares
pinv
 Pseudoinverse
lscov
Least squares with known covariance
The inverse of a square


If [A] is a square matrix, there is another matrix [A]-1,
called the inverse of [A], for which [A][A]-1=[A]-1[A]=[I]
The inverse can be computed in a column by column
fashion by generating solutions with unit vectors as the
right-hand-side constants:

1

Ax1  0

0


0

Ax 2   1

0

A  x1
1

x2

0

Ax 3  0

1

x3 
Canonical base of an n-dimensional vector space
100……000
010……000
001……000
…………….
000…….100
000…….010
000…….001
Matrix Inverse (cont)

LU factorization can be used to efficiently evaluate a
system for multiple right-hand-side vectors - thus, it is
ideal for evaluating the multiple unit vectors needed to
compute the inverse.
The response of a linear system

The response of a linear system to some stimuli can be found
using the matrix inverse.
Interactions response   stimuli 
Ar  s

1
1
A Ar  A s
1
A A I
1
rA s
Distance and norms

Metric space  a set where the ”distance” between elements of the
set is defined, e.g., the 3-dimensional Euclidean space. The
Euclidean metric defines the distance between two points as the
length of the straight line connecting them.

A norm  real-valued function that provides a measure of the size
or “length” of an element of a vector space.
Vector Norms


The p-norm of a vector X is:
X
p
1/ p
 n

p
  x i 
i1

Important examples of vector p-norms include:

p  1:sum of the absolute values
n
X 1   xi
i1
p  2 : Euclidian norm (length)
p   : maximum  magnitude
X2 X e
X

 max xi
1in
n
x
2
i
i1
Matrix Norms

Common matrix norms for a matrix [A] include:
n
column - sum norm
Frobenius norm
A 1  max  aij
1 jn
A
f

n
i1
n
 a
i1 j1
n
row - sum norm
spectral norm (2 norm)
A
A

2
 max  aij
1in
j1
1/2
  max 
Note - max is the largest eigenvalue of [A]T[A].


2
ij
Matrix Condition Number




The matrix condition number Cond[A] is obtained by calculating
Cond[A]=||A||·||A-1||
In can be shown that:
X
A
 Cond A
X
A
The relative error of the norm of the computed solution can be as
large as the relative error of the norm of the coefficients of [A]
multiplied by the condition number.

If the coefficients
of [A] are known to t digit precision, the solution [X]
may be valid to only
t-log10(Cond[A]) digits.
Built-in functions to compute norms and
condition numbers



norm(X,p)  Compute the p norm of vector X, where p
can be any number, inf, or ‘fro’ (for the Euclidean norm)
norm(A,p)  Compute a norm of matrix A, where p can
be 1, 2, inf, or ‘fro’ (for the Frobenius norm)
cond(X,p) or cond(A,p)  Calculate the condition
number of vector X or matrix A using the norm specified
by p.
LU Factorization

LU factorization involves two steps:

Decompose the [A] matrix into a product of:



a lower triangular matrix [L] with 1 for each entry on the diagonal.
and an upper triangular matrix [U
Substitution to solve for {x}

Gauss elimination can be implemented using LU factorization

The forward-elimination step of Gauss elimination comprises the
bulk of the computational effort.
LU factorization methods separate the time-consuming elimination
of the matrix [A] from the manipulations of the right-hand-side [b].

Gauss Elimination as LU Factorization







To solve [A]{x}={b}, first decompose [A] to get [L][U]{x}={b}
MATLAB’s lu function can be used to generate the [L] and [U] matrices:
[L, U] = lu(A)
Step 1  solve [L]{y}={b}; {y} can be found using forward substitution.
Step 2 solve [U]{x}={y}, {x} can be found using backward substitution.
In MATLAB:
[L, U] = lu(A)
d = L\b
x = U\d
LU factorization  requires the same number of floating point operations
(flops) as for Gauss elimination.
Advantage  once [A] is decomposed, the same [L] and [U] can be
used for multiple {b} vectors.
Cholesky Factorization





A symmetric matrix  a square matrix, A, that is equal to
its transpose:
A = AT (T stands for transpose).
The Cholesky factorization  based on the fact that a symmetric
matrix can be decomposed as:
[A]= [U]T[U]
The rest of the process is similar to LU decomposition and Gauss
elimination, except only one matrix, [U], needs to be stored.
Cholesky factorization with the built-in chol command:
U = chol(A)
MATLAB’s left division operator \ examines the system to see which
method will most efficiently solve the problem. This includes trying
banded solvers, back and forward substitutions, Cholesky
factorization for symmetric systems. If these do not work and the
system is square, Gauss elimination with partial pivoting is used.
Download